I'm trying to implement a simple OR calculating perceptron with two inputs and a single output. Unfortunately, although the weights and bias do converge on particular values, they aren't the right ones, and the values they converge on depend on what order inputs are given in. The process I use to train the perceptron looks like this: The threshold equals input 1 * weight 1 + input 2 * weight 2 + bias If the threshold > 0, the answer is 1 Otherwise, the answer is 0 The training looks like this: After calculating the threshold, the weights are modified as follows: New weight value = old weight value + corresponding input value * learning factor * (desired value - threshold value) New bias value = old bias value + (desired value - threshold value) When I input all valid combinations [0, 0][1, 0][0, 1][1, 1] in one order, the weights and bias settle out to the values of 1/3, 1/3, 2/3, or 1/3, 1/3, 0, neither of which give the correct answer in all scenarios. Can you spot my error? Thanks a bunch!